"Do the hard 20% manually, let LLMs handle the rest": Naval boosts viral coding take
naval · x · 2026-09-26
Naval reposted @rygo6's framework for using LLMs in software development:
- LLMs can get you 80% of the way upfront, but the last 20% becomes nearly impossible to overcome with prompting alone
- The right approach is inverted: manually and deliberately build the hardest 20% first — architecture, API surface, data structures, base libraries
- That manually built 20% becomes part of your development harness, almost a prompt itself, letting you prompt your way through the remaining 80%
The quoted thread adds a concern: knowing what belongs in that first 20% requires staff/principal-level domain knowledge — LLMs may actually raise the value of senior engineers.
More from coding & agent
- Demo shows humans taking over AI agents mid-task on a real computer — aniketmaurya · 2026-09-26
- You.com uses OpenRouter as reranking layer: 3x fewer tokens, 84% benchmark accuracy — RichardSocher · 2026-09-26
- Microsoft standardizes on GitHub Copilot SDK across products, ex-engineer reveals — martinwoodward · 2026-09-26
- Creator builds 3D model in ComfyUI, audio-reactive stage with Sentinel — PurzBeats · 2026-09-26
- OOD Labs launches Sentinel, a $199.99 real-time node graph built for AI agents — PurzBeats · 2026-09-26
- Dev builds Litmus, a personalized AI-writing detector, tests 7 models — Sol sounds most human — chaseleantj · 2026-09-26